[2/4] alfred: Allow operating without any interface specified

  The '-i' commandline parameter to specify interface names
no longer is mandatory. Specifying interface 'none' or
sending a 'none' interface string within the
ALFRED_CHANGE_INTERFACE unix socket command disables
all interfaces operations at runtime.

Signed-off-by: Marek Lindner <mareklindner@neomailbox.ch>
---
 README.rst   |  6 +++++-
 alfred.h     |  1 -
 man/alfred.8 |  6 +++++-
 netsock.c    | 23 ++++-------------------
 server.c     | 12 ------------
 5 files changed, 14 insertions(+), 34 deletions(-)
  

On Sunday, 2 January 2022 12:31:34 CET Marek Lindner wrote:
[...]
> diff --git a/README.rst b/README.rst
[...]
>  	if (strcmp(globals->mesh_iface, "none") != 0 &&
>  	    batadv_interface_check(globals->mesh_iface) < 0 &&
>  	    !globals->force) {
> @@ -393,12 +387,6 @@ int alfred_server(struct globals *globals)
>  		return -1;
>  	}
>  
> -	num_socks = netsock_open_all(globals);
> -	if (num_socks <= 0 && !globals->force) {
> -		fprintf(stderr, "Failed to open interfaces\n");
> -		return -1;
> -	}
> -
>  	num_interfaces = netsocket_count_interfaces(globals);
>  	if (num_interfaces > 1 && globals->opmode == OPMODE_SECONDARY) {
>  		fprintf(stderr, "More than one interface specified in secondary mode\n");
> 

This now causes the "--force" option (+its storage in the globals data 
structure) to be completely useless. I would prefer to have this handling
still be there when !list_empty(&globals->interfaces).

And why is it necessary to not open the sockets on startup when interfaces are 
already given?

Kind regards,
	Sven
  

On Sunday, 2 January 2022 15:20:20 CET Sven Eckelmann wrote:
> This now causes the "--force" option (+its storage in the globals data
> structure) to be completely useless. 

Why would global->force be useless ? The alfred_server() function still uses 
the global->force state to determine if globals->mesh_iface is configured 
correctly.


> I would prefer to have this handling still be there when
> !list_empty(&globals->interfaces).

To be honest, I hadn't fully understood what use case global->force is trying 
to address. What do you have in mind ? Checking for list_empty() will require 
alfred to be always started with an interface configured while alfred could be 
used without any interface at all and operate as local data storage between 2 
processes on the same system or the interface could be configured at a later 
time (via unix socket).


> And why is it necessary to not open the sockets on startup when interfaces
> are already given?

The main while loop calls netsock_reopen() in each round which will open all 
necessary sockets (unless I am mistaken). My guess is that this was added when 
the ALFRED_CHANGE_INTERFACE call was added. Therefore, the netsock_open() call 
is somewhat redundant unless alfred is meant to always require an interface at 
startup time and alfred is meant to bail out whenever that configured interface 
isn't available at startup time.

Cheers,
Marek
  

On Sunday, 2 January 2022 20:01:47 CET Marek Lindner wrote:
> On Sunday, 2 January 2022 15:20:20 CET Sven Eckelmann wrote:
> > This now causes the "--force" option (+its storage in the globals data
> > structure) to be completely useless. 
> 
> Why would global->force be useless ? The alfred_server() function still uses 
> the global->force state to determine if globals->mesh_iface is configured 
> correctly.

Ok, you are right about mesh_iface. But the patch missed to adjust the manpage 
to clarify this.

> > I would prefer to have this handling still be there when
> > !list_empty(&globals->interfaces).
> 
> To be honest, I hadn't fully understood what use case global->force is trying 
> to address. What do you have in mind ? Checking for list_empty() will require 
> alfred to be always started with an interface configured while alfred could be 
> used without any interface at all and operate as local data storage between 2 
> processes on the same system or the interface could be configured at a later 
> time (via unix socket).

No, I wasn't talking about list_empty() but about !list_empty(). You removed 
the first block because you want to have have alfred started without any 
interface - fine with that.

But the default behavior of alfred in the past was to first check if the 
selected interfaces make sense and then return an error if there was a problem 
to open them. The force option basically ignored any error when there was an 
interface not ready yet. But the patch completely removed the chance to pre-
check the interfaces on startup.

> > And why is it necessary to not open the sockets on startup when interfaces
> > are already given?
> 
> The main while loop calls netsock_reopen() in each round which will open all 
> necessary sockets (unless I am mistaken). My guess is that this was added when 
> the ALFRED_CHANGE_INTERFACE call was added. Therefore, the netsock_open() call 
> is somewhat redundant unless alfred is meant to always require an interface at 
> startup time and alfred is meant to bail out whenever that configured interface 
> isn't available at startup time.

See above. The situation I had in mind was:

* force not enabled
* an interface given to the process
* interface cannot be used by alfred

Kind regards,
	Sven